Linear semi-openness and the Lyusternik theorem
Introduction
A classical theorem of Lyusternik [11], modified later by Graves [6], states that the tangent space to a level set of a C1-mapping F (between Banach spaces) at a point x̄ is the kernel of the Fréchet derivative ––provided the latter is surjective. This hypothesis implies, by Banach's open mapping theorem, that the continuous linear mapping is open, which is the crucial property for verifying the statement.
By now, numerous modifications and generalizations of the Lyusternik theorem have been established, where the “derivative” is no longer assumed to be an open linear mapping (see, e.g., [2], [5], [10], [13], [14]). In this context, it turned out that the concept of linear openness (or openness at a linear rate) is an appropriate tool. This concept as well as the closely related metric regularity have been investigated in a lot of papers (we refer in the following to [7], [8], [12], [15]).
The aim of the present paper is to derive a Lyusternik type theorem for a class of set-valued mappings called here linearly semi-open mappings (Section 3). This new concept is equivalently characterized by notions that have their analogies in metric regularity and the pseudo-Lipschitz property (Section 2). Furthermore, Section 2 contains a perturbation result for linear semi-openness. The mappings that are open at a linear rate form a subclass of the linearly semi-open mappings. To provide further examples for linear semi-open mappings, linear processes are investigated (Section 4). In this connection, the computation of openness bounds of linear open mappings in the classical sense gains importance; some results concerning this problem are presented. Finally, conditions are given which ensure that a continuous piecewise linear function is linearly semi-open (Section 5).
Throughout the paper we use the following notations. X and Y are real Banach spaces, BX and BY, or simply B, denotes the respective closed unit ball in the space.
Let be a set-valued mapping of X into Y, i.e., a mapping of X into the power set of Y. The set is called domain of F. The graph of F is the set gr(F)={(x,y)∈X×Y:y∈F(x)}. We define the kernel of F as the set . The inverse mapping is defined by F−1(y)={x∈X:y∈F(x)}. We identify the (single-valued) mapping F:X→Y with the set-valued mapping defined by , x∈X.
The closure, the interior, and the convex hull of a subset M of a Banach space are denoted by cl(M), int(M), and co(M), respectively.
A nonempty subset M of a Banach space is said to be a cone, if x∈M and λ⩾0 imply λx∈M. The cone generated by a nonempty set S is denoted by cone(S).
Section snippets
The concept
We investigate a mapping . Let be given. Definition 1 F is said to be linearly semi-open around if there are numbers c>0 and t0>0 such thatis satisfied for all (x,y)∈gr(F) with , and all t∈[0,t0]. The numbers c and t0 are said to be semi-openness parameters. The openness parameter c is sometimes referred to as openness rate.
Each linearly open mapping is linearly semi-open. Remember that a mapping is said to be linearly open (or open
A Lyusternik type theorem
Metric regularity, as a description of a certain stability of the inverse mapping, has been used to derive generalized Lyusternik type theorems (see for example [1], [14]). In the following section, we show that for this purpose the assumption of metric regularity can be weakened.
We define the contingent cone to a set M⊂X at a point in the following way:Moreover, for we consider the mapping , which assigns to every
Uniform openness parameters
To bring linear semi-openness and especially Theorem 3 into use, it is necessary to find classes of mappings that actually enjoy the property of being linearly semi-open. To this aim we now consider processes, i.e., set-valued mappings the graphs of which are cones. Then, the stability result given in Section 2.3 can be used to derive linear semi-openness of other mappings too. Recall that a process is said to be convex or closed if its graph has the respective property. A process is called
Piecewise linear functions
We now turn to a special class of processes. Let be a continuous piecewise linear function. This means: F is continuous, and there are finitely many linear selection functionals such that for each point there is an index i∈{1,…,p} satisfying F(x)=Aix. Define the following sets:
Mi={x:F(x)=Aix},
I(x)={i:x∈Mi} the active index set, and
the essential active index set (cf. [9]).
Notice that F is a process and is globally Lipschitz on . We calculate the
Acknowledgements
The authors would like to thank the referees for their valuable suggestions.
References (15)
- et al.
Set-Valued Analysis
(1990) Stability and regular points of inequality systems
Journal of Optimization Theory and Applications
(1986)- et al.
Verifiable necessary and sufficient conditions for openness and regularity of set-valued and single-valued maps
Journal of Mathematical Analysis and Applications
(1988) Optimization and Nonsmooth Analysis
(1983)Metric regularity, tangent sets, and second-order optimality conditions
Applied Mathematics and Optimization
(1990)Some mapping theorems
Duke Mathematical Journal
(1950)- et al.
Metric regularity for strongly compactly lipschitzian mappings
Nonlinear Analysis
(1995)